Reproduce “SM4RT: Cascaded Feed-forward Model for 4D Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pang-Chi Lo, Jhao-Syun Lai, Tsung-Wei Ke
SM4RT estimates 3D scene geometry and motion from two video frames. It first predicts point maps, then uses their 3D positions to guide object-flow prediction, measuring each pixel’s displacement in its input frame’s camera coordinates. On held-out Waymo clips, its forward EPE is 0.013 versus DPM’s 0.020; its geometry results are competitive, not uniformly best.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.